Ubuntu 16.04, CUDA 8 - CUDA driver version is insufficient for CUDA runtime version

前端 未结 9 734
孤街浪徒
孤街浪徒 2020-12-08 04:15

I\'ve installed the latest nvidia drivers (375.26) manually, and installed CUDA using cuda_8.0.44_linux.run (skipping the driver install there, since the bundled drivers are

相关标签:
9条回答
  • 2020-12-08 05:05

    I had similar problem.I am using anaconda, after installing keras-gpu through conda, it automatically took the most recent version of cuda, which was not compatible with my system.

    You can view list of installed packages in anaconda through

    conda list
    

    In my case supported version was 10.0 but anaconda installed the latest version 10.1. if this is the case, you need to reinstall cuda in anaconda enviournment with supported version.

    0 讨论(0)
  • 2020-12-08 05:10

    I got that error on Ubuntu 16.04, because I was still using the open-source X.Org video driver. The error went away when I switched to the NVIDIA binary driver.

    I found the driver settings by opening the System Settings, then clicking on Software & Updates. The video drivers are on the Additional Drivers tab.

    0 讨论(0)
  • 2020-12-08 05:11

    I was with the same problem. I had the version nvidia-390 installed on Ubuntu 18.04.2 LTS. My Graphic card is GeForce GTX 1080, and using tensorflow 1.12.0. I successfully solved this problem by removing the old version:

    sudo apt-get purge nvidia-*

    And then installing the version 418

    sudo apt-get install nvidia-driver-418 nvidia-settings

    0 讨论(0)
提交回复
热议问题